post: a current station's coordinate is a label, not the hydraulic control

[clarkbw]

Post: 2026-08-26-current-station-coordinate-is-a-label-not-the-hydraulic-control-…md

Draft post from the Dodd Narrows hydraulic-control fix — station-corrections PR #15, released in v2.9.2.

Title: A current station's coordinate is a label, not the hydraulic control
Date is the draft date (2026-08-26) — script/publish.py bumps it at merge.

The angle

Published reference-data coordinates are identifiers, not measurements. The CHS current station for Dodd Narrows sits 120 m from the pass's 80 m throat, on water with 1.9× the cross-sectional area — a rounding error on a chart, a factor-of-two error in a continuity model. The generalizable lesson is the one the post leads with: fix it once in shared reference data, with a source: field, and land it separately from the feature that found it.

Claims verified against the repos

Claim	Status
CHS/IWLS publishes Dodd at 49.134351 / -123.817132	verified live against the IWLS API
Recovered control coordinate 49.13546639419797 / -123.81735084108287, source: string, v2.9.2	verified on origin/main
Anchor section 160 m / 2322 m²; gut 80 m / 1216 m²; 9.43 × 2322/1216 = 18.01 kn (+91 %)	verified, arithmetic re-checked
Anchor stability 19.91 / 20.52 / 19.67 % vs a 10 % bar	verified
100 m re-run is worse: 27.62 / 33.64 / 20.33 %	verified
Stage interpolation error 0.9164 % at withheld half-metre heights	verified
Mesh reads 1.46 kn where truth is 6.2 kn	verified

Corrections to the brief I was given

• The offset is 120.0 m, not 130 m. 130 m is the pipeline's section-centre-to-section-centre figure; 120.0 m (bearing 351°) is the measured distance between the two positions the registry actually ships. The post explains the difference rather than papering over it.
• Moving the anchor did not clear the 10 % bar. It fixed the two section-placement variants (19.9/20.5 % → 8.6/9.6 %) and made the chart-datum variant worse (19.7 % → 22.67 %). Position error and datum sensitivity turned out to be two separate problems that summed to one failing number. That's now the strongest section of the post.
• Dodd still ships no current patch. Even with a modelled stage-dependent area (which itself certifies fine), the surviving range collapses to a single transect and packs zero cells; the NONNA-10 fallback fails too. The post ends on that honestly — the coordinate correction outlived the pipeline that found it, which is the actual thesis.
• Not everything downstream got it "for free." Three readers, three cadences: npm bump / build-time generator re-run / vendored copy guarded by a drift test. Figure 3 shows this.

Figures (3, hand-written SVG)

1. anchor-stability-before-after.svg — the three sensitivity variants before/after the anchor move, against the 10 % bar. Shows the datum axis getting worse.
2. dodd-anchor-vs-control.svg — plan view: published coordinate vs the 80 m control, 120 m apart, with the continuity arithmetic.
3. registry-one-edit-three-readers.svg — one authoring edit, three pull cadences.

Privacy / policy

• No provider-minted station code anywhere in the post (the curl example elides it) — consistent with the registry's own posture and the published licensing post.
• No private repo named. slackwater-ios appears only as "an offline tide and current app"; the patch pipeline is referred to by script name, never by repo.
• No roadmap, milestone, or GTM content.
• GSC DEM (Open File 8963, OGL – Canada) and CHS current predictions are linked.

Verified before opening

• python3 script/lint-liquid.py → 42 posts clean
• bundle exec jekyll build → clean
• _site/<slug>/index.html renders all three /assets/img/… figures; SVGs land in _site/assets/img/<slug>/; all three post_url related links resolve
• Privacy grep for hostnames / IPs / local paths across post + SVGs → nothing